pam-cdr

This code is designed offer simulations to help one understand and model high speed PAM-N systems (specifically PCIe systems) and is also supplemental material to my thesis titled Study and simulation of PCIe 6, with an emphasis on the physical layer.

The impulse_responses directory contains impulses responses sampled at 128GSa/s of realistic channels from the PCIe 6.0 standard, and include a low loss channel and a normal channel and example impulse responses of an aggressor (ie an adjacent cross talk lane). These impulse responses came from extracting s-parameters from Keysight ADS on differential pairs and then combining them appropriately to represent a response that effectively involes going from a single ended signal through to a differential channel and then being converted back to a single ended signal.

To run these examples its required to have Python3 or above, and the packages matplotlib, numpy, scipy, and scikit-dsp-comm are requred.

• initial_cdr investigates performing clock data recovery with either an early late or a MM TED and a second order loop filter on a relatively ideal, raised cosine pulse shape and supports a wide variety of environment variables to alter its behavior.
• ctle.py generates the bilinear transformed poles and zeros of the Continuous Time Linear Equalizer as described by the PCIe 6.0 standard
• test_equalization.py and test_equalization_low_loss.py run an LMS algorithm to determine DD FFE and DFE coefficients for the medium and low loss channels respectively and support a wide variety of environment variables. Note that at 4 Samples/Symbol and PAM-4 the medium loss channel need Aopt to be initialized from coefficients determined from running a PAM-2 simulation. Here hand-of-God timing synchronization is done through the group_delay_fude_factor variable (which may need to tuned for different filter choices). There are also options to add an aggressor channel into the system.
• aggregate_system.py and aggregate_system_low_loss.py run simultaneous timing recovery and LMS equalization on a receiver for the medium and low loss channels respectively and support a wide variety of environment variables. This shows the complex dynamics of performing simultaneous channel equalization and timing estimations. These simulations are oftentimes aided by providing initial tap weights Aopt from a run of the test_equalization simulation to aid in simultaneous convergence.